EP0506327A2 - A system and method for ranking and extracting salient contours for target recognition - Google Patents
A system and method for ranking and extracting salient contours for target recognition Download PDFInfo
- Publication number
- EP0506327A2 EP0506327A2 EP92302501A EP92302501A EP0506327A2 EP 0506327 A2 EP0506327 A2 EP 0506327A2 EP 92302501 A EP92302501 A EP 92302501A EP 92302501 A EP92302501 A EP 92302501A EP 0506327 A2 EP0506327 A2 EP 0506327A2
- Authority
- EP
- European Patent Office
- Prior art keywords
- contours
- saliency
- contour
- curvature
- max
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/12—Edge-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/181—Segmentation; Edge detection involving edge growing; involving edge linking
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/255—Detecting or recognising potential candidate objects based on visual cues, e.g. shapes
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
- G06V10/443—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components by matching or filtering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
- G06V10/462—Salient features, e.g. scale invariant feature transforms [SIFT]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10016—Video; Image sequence
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10032—Satellite or aerial image; Remote sensing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10048—Infrared image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20016—Hierarchical, coarse-to-fine, multiscale or multiresolution image processing; Pyramid transform
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30212—Military
Definitions
- This invention relates generally to pattern recognition, and more specifically to a system and method for ranking and extracting salient contours for improved target recognition.
- ATR automatic target recognition
- Edge detectors have been developed to attempt to obtain such robust descriptions. These detectors usually have thresholds to control which edges they find in the image. For example, the Canny edge detector has two thresholds controlling which edges are retained and which are thrown away. However, because the thresholds in the Canny edge detector are based on just the strengths of the edges, the resulting edges are not salient contours corresponding to object boundaries. On the other hand, desired object boundaries may not necessarily be strong but may have other important characteristics.
- a method recently developed extracts salient axes of symmetry called a skeleton sketch, although the method does not work with real images and does not find salient contours.
- Another method using indoor images employs a perceptual organization for grouping edges based on colinearity, parallelism, and co-termination. Such a method is not robust enough for outdoor images.
- an object of the present invention to provide a system and method for achieving robust and efficient object recognition, particularly from real outdoor images by reducing clutter and extracting salient information about objects in a complex image.
- Each linked edgel (contour) is assigned a salient measure for each of length, smoothness, and contrast.
- the number of edgels in a contour is used for length, the average change of curvature is used for smoothness, and edge magnitude is used for contrast.
- a saliency value for each of the contours is computed based on the saliency measures for the respective contour, the contours are ranked in decreasing order of saliency, and certain ones of the ranked contours (which correspond to the object of interest) are selected based on the requirements of the particular vision application in which they will be used.
- a contour is defined as a linked list of edge pixels.
- a salient contour therefore, is one that is more significant than others in an image, wherein significance is established by the importance of certain characteristics of a contour for reliably finding objects of interest (referred to as targets) in the image.
- Salient contours are important for various applications such as object recognition and tracking of objects in moving images, discussed later.
- Extracting salient contours decreases image clutter.
- there is more information about the target relative to the background thereby making subsequent processing, especially object recognition and matching, more robust and efficient in time and memory.
- a system and method for salient contour extraction according to the present invention is quite useful in a number of vision applications.
- a contour is considered salient if it is long, or smooth, or contrasts strongly with the surrounding region (this is an inclusive or). According to a preferred embodiment of the present invention, each one of these characteristics have been assigned a saliency measure. These saliency measures are then combined into one value for each contour.
- a preferred embodiment of the salient contours extraction method according to the present invention has been implemented in Common Lisp on a Texas Instruments Explorer II.
- the time complexity is O(n), where n is the number of edgels in the image.
- the present invention should not be limited to any particular language or computing platform, as its performance is not particularly affected by any specific implementation selected.
- the systems and methods of the present invention can be easily implemented in parallel, because the saliency attributes (length, smoothness, and contrast) are independent of one another and independent from contour to contour.
- Contours corresponding to three-dimensional phenomena like depth discontinuities and orientation discontinuities are preferred over those corresponding to two-dimensional phenomena like illumination changes in the world (e.g., shadows, surface, and reflectance markings). Since the input received from the sensor is two-dimensional, these preferences must be made from two-dimensional data.
- FIG. 1 is a block diagram of a vision system 10 embodying the present invention
- sensor 15 which is preferrably a passive sensor, although this is not required.
- Edge detector 20 such as a Canny edge detector (although other edge detectors could be used), receives the image from sensor 15 and passes over the image to detect edge pixels.
- the edge pixels also known as edgels
- contours are output from edge detector 20 and are linked into lists of edgels called contours by linker 25.
- saliency value extractor 30 which comprises saliency measure estimators 35,40,45,50,55 where a saliency measure for each desired characteristic for each contour is computed, and saliency value ranking unit 60 where all saliency measures for each contour are combined to obtain a saliency value for that contour.
- the average contrast of a contour is the average of the strength of each edgel in the contour.
- the strength of each edgel is obtained from edge detector 20.
- the saliency measures for each characteristic for a particular contour is then received by saliency value ranking unit 60.
- ⁇ l is the weight for the length saliency measure
- ⁇ s is the weight for the smoothness saliency measure
- ⁇ c is the weight for the contrast saliency measure.
- the contours are also ranked in decreasing order of saliency by saliency value ranking unit 60.
- the most salient contours are then selected to be provided to vision application 65, which could be, for example, an object matching system.
- Figures 2a-e involve an image of a flat bed truck obtained by a long wave forward-looking infra-red (FLIR) sensor
- Figures 3a-e relate to an image of a small communications van obtained in the visible spectrum by a TV sensor. Discussion about Figures 3a-e is identical to that which follows for Figures 2a-e.
- FLIR long wave forward-looking infra-red
- Figure 2a shows the gray level images received from passive sensor 20.
- Figure 2b shows the Canny edges (edgels) obtained at small thresholds of edge detector 20 to give a large number of edges.
- Figure 2c shows the most salient contours obtained from saliency value extractor 30.
- FIG. 2d shows the cumulative number of on-target and off-target corners for the 50 most-salient contours. It can be seen that both the on-target and off-target curves are monotonically nondecreasing because they are cumulative curves. For better operation of vision application 65, it is preferable to have more on-target corners and less off-target corners in the top few contours. Therefore, especially in the most salient contours, it is desirable to have the on-target curve above the off-target curve, with a wide gap between the two curves.
- the bar chart ( Figure 4), corresponding to the gray level images of Figures 2a and 3a, shows the increase in the fraction of target corners obtained when only the most salient contours extracted according to the present invention are used.
- the darker bars show the fraction of target corners ascertained without saliency value extractor 30, while the lighter bars reflect the fraction of target corners established by employing saliency value extractor 30.
- the present invention is useful for a number of computer vision tasks, including object recognition.
- object recognition By separating an object from the background within an image, salient contours extraction becomes an important precursor to object recognition. Furthermore, since the clutter has been removed to a large extent, an object recognition system can proceed faster and more efficiently (i.e., more robust recognition).
- an object recognition application of vision system 65 matches two-dimensional corners in an image, received from saliency value combination unit 60, with the vertices of a three-dimensional model using an alignment technique.
- Object recognition system 65 first extracts corners in the image provided by saliency value ranking unit 60. Then every triple of image corners is matched to every triple of three-dimensional model vertices to generate an alignment. The alignment is used to project the three-dimensional model to the two-dimensional image for verification.
- the salient contour extraction system and method of the present invention increases the ratio of image corners on the object to all image corners, thereby reducing the number of extraneous corners caused by image clutter, and resulting in a reduction of the search space (the combinations) in the corner matching stage of object recognition.
- Salient contours are also employed in a verification stage where only the salient image contours are used to verify the projected model contours.
- FIG. 5 In motion matching situations ( Figure 5), important for tracking objects within images, a plurality of image frames F1,F2,F3,...F n from a motion sequence are provided to vision system 10 embodying the present invention. Vision system 10 extracts the most salient contours from frames F1,F2,F3,...F n and provides these contours to two dimensional matcher vision application 65. It should be noted that S1,S2,S3,...S n may comprise one vision system 10 or a plurality of vision systems 10 operating in parallel. Matcher 65 then matches these contours to estimate motion parameters. The use of extracted salient contours makes it considerably easier for matcher 65 to match edges in frames of a motion sequence. Matcher 65 may then yield two-dimensional or three-dimensional motion estimation based on inferencing after the matching operation, may determine structure of the object of interest from its motion, may operate as a moving target indicator, etc.
- FIG 6 illustrates an application of the present invention in a stereo matching situation
- vision system 10 extracts the most salient contours from frames F1,F2 and provides these contours to matcher vision application 65.
- S1,S2 may comprise one vision system 10 or a plurality of vision systems 10 operating in parallel.
- Matcher 65 then matches these contours to estimate depth and other three-dimensional information about the scene, based on inferencing. As in the previous matching situation, extracting salient contours according to the present invention makes matching much easier since clutter in the image is significantly reduced.
- the results achieved by use of the present invention are fairly independent of the kind of two-dimensional sensor 15 used (e.g., TV, infra-red, etc.). Furthermore, the present invention operates primarily in a local context especially because an image may be occluded and/or there could be long, thin objects in the image.
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
Description
- This invention relates generally to pattern recognition, and more specifically to a system and method for ranking and extracting salient contours for improved target recognition.
- The central problem in machine vision and automatic target recognition (ATR) is to obtain robust descriptions from images. Robust descriptions help not only in object recognition, but also in motion matching, stereo matching, as well as in generally reducing image clutter. For object recognition, in particular, good descriptions increase the probability of recognition and reduce the probability of false alarms.
- Edge detectors have been developed to attempt to obtain such robust descriptions. These detectors usually have thresholds to control which edges they find in the image. For example, the Canny edge detector has two thresholds controlling which edges are retained and which are thrown away. However, because the thresholds in the Canny edge detector are based on just the strengths of the edges, the resulting edges are not salient contours corresponding to object boundaries. On the other hand, desired object boundaries may not necessarily be strong but may have other important characteristics.
- Other systems have been developed which employ total curvature as well as total curvature variation. Such systems obviously prefer circles and edges with substantial curvature. Additionally, such systems interatively converge to a solution and require a network to compute saliency.
- A method recently developed extracts salient axes of symmetry called a skeleton sketch, although the method does not work with real images and does not find salient contours. Another method using indoor images (which have significantly less clutter and imperfection compared to outdoor images) employs a perceptual organization for grouping edges based on colinearity, parallelism, and co-termination. Such a method is not robust enough for outdoor images.
- Other related research involves methods for intensity-based region segmentation. These techniques group image pixels into regions based on some similiarity measure. There are several problems with these methods, such as the fact they always yield closed boundaries. In complex, real outdoor images closed boundaries usually correspond to highlights, shadows, etc., but rarely correspond to object boundaries. Another problem is that the region boundaries are extremely sensitive to thresholds, (i.e., such boundaries change and move significantly as thresholds change). Yet another problem is that region segmentation is a global operation, meaning it is less sensitive and more likely to miss low contrast boundaries or small objects, particularly in complex images. For example, long, thin objects may be missed, although they may be salient.
- Accordingly, improvements which overcome any or all of these problems are presently desirable.
- In view of the above problems associated with the related art, it is an object of the present invention to provide a system and method for achieving robust and efficient object recognition, particularly from real outdoor images by reducing clutter and extracting salient information about objects in a complex image.
- It is another object of the present invention to provide a system and method for ranking and extracting salient contours corresponding to possible objects from an image edge map or from a two-dimensional image acquired by a passive sensor.
- These and other objects are accomplished in a preferred embodiment of the present invention by a system and method for obtaining salient contours from two-dimensional images acquired by a sensor which process the two-dimensional images with an edge detector to produce edgels from each of the images and link the edgels into lists known as contours. Each linked edgel (contour) is assigned a salient measure for each of length, smoothness, and contrast. The number of edgels in a contour is used for length, the average change of curvature is used for smoothness, and edge magnitude is used for contrast. A saliency value for each of the contours is computed based on the saliency measures for the respective contour, the contours are ranked in decreasing order of saliency, and certain ones of the ranked contours (which correspond to the object of interest) are selected based on the requirements of the particular vision application in which they will be used.
- These and other features and advantages of the invention will be apparent to those skilled in the art from the following detailed description of a preferred embodiment, taken together with the accompanying drawings, in which:
-
- FIG. 1 is a block diagram of a vision system according to a preferred embodiment of the present invention;
- FIGs. 2a-e are pictorial descriptions and charts demonstrating how the present invention affects an image obtained by a long wave forward-looking infra-red (FLIR) sensor;
- FIGs. 3a-e are pictorial descriptions and charts illustrating how the present invention affects an image obtained in the visible spectrum by a TV sensor;
- FIG. 4 is a bar chart for the gray level images of FIGs. 2a and 3a, showing the increase in fraction of target corners obtained achieved by employing the present invention;
- FIG. 5 illustrates an application of the present invention in a motion matching situation; and
- FIG. 6 illustrates an application of the present invention in a stereo matching situation.
- Corresponding numerals and symbols in the different figures refer to corresponding parts unless otherwise indicated.
- A contour is defined as a linked list of edge pixels. A salient contour, therefore, is one that is more significant than others in an image, wherein significance is established by the importance of certain characteristics of a contour for reliably finding objects of interest (referred to as targets) in the image. Salient contours are important for various applications such as object recognition and tracking of objects in moving images, discussed later.
- Extracting salient contours decreases image clutter. Thus, there is more information about the target relative to the background, thereby making subsequent processing, especially object recognition and matching, more robust and efficient in time and memory. For this reason, a system and method for salient contour extraction according to the present invention is quite useful in a number of vision applications.
- It is useful to first characterize salient contours in terms of their physical attributes and then describe preferred embodiments of the present invention.
- A contour is considered salient if it is long, or smooth, or contrasts strongly with the surrounding region (this is an inclusive or). According to a preferred embodiment of the present invention, each one of these characteristics have been assigned a saliency measure. These saliency measures are then combined into one value for each contour.
- A preferred embodiment of the salient contours extraction method according to the present invention has been implemented in Common Lisp on a Texas Instruments Explorer II. The time complexity is O(n), where n is the number of edgels in the image. It should be realized that the present invention should not be limited to any particular language or computing platform, as its performance is not particularly affected by any specific implementation selected. Note too, the systems and methods of the present invention can be easily implemented in parallel, because the saliency attributes (length, smoothness, and contrast) are independent of one another and independent from contour to contour.
- Contours corresponding to three-dimensional phenomena like depth discontinuities and orientation discontinuities are preferred over those corresponding to two-dimensional phenomena like illumination changes in the world (e.g., shadows, surface, and reflectance markings). Since the input received from the sensor is two-dimensional, these preferences must be made from two-dimensional data.
- Beginning with Figure 1, which is a block diagram of a
vision system 10 embodying the present invention, an image is sensed bysensor 15 which is preferrably a passive sensor, although this is not required.Edge detector 20, such as a Canny edge detector (although other edge detectors could be used), receives the image fromsensor 15 and passes over the image to detect edge pixels. Next, the edge pixels (also known as edgels) are output fromedge detector 20 and are linked into lists of edgels called contours by linker 25. These contours are provided tosaliency value extractor 30 which comprisessaliency measure estimators value ranking unit 60 where all saliency measures for each contour are combined to obtain a saliency value for that contour. Within lengthsaliency measure estimator 35, the length saliency measure, S l, of a contour is given by
where L, the length of the contour, is defined as the number of edgels in that contour, Lmin is the length of the shortest contour, and Lmax, is the length of the longest contour. -
-
- 1. Compute curvature at each point in the contour, preferably using B-spline fitting.
- 2. Compute the change of curvature at each point by convolving the curvature values using the difference operator:
- 3. Compute the average of the change of curvature.
-
- The average contrast of a contour is the average of the strength of each edgel in the contour. The strength of each edgel is obtained from
edge detector 20. - Although only three characteristics are discussed in connection with the particular embodiment of the present invention being discussed, it should be understood that, as shown in Figure 1, there could be additional saliency characteristics (
saliency measure estimators - The saliency measures for each characteristic for a particular contour is then received by saliency
value ranking unit 60. The saliency value, S, of a contour is then computed as
where ωl is the weight for the length saliency measure, ωs is the weight for the smoothness saliency measure, and ωc is the weight for the contrast saliency measure. - The weights for l, s, and c (and others if desired) are preselected based on the importance of the various measures; the higher the saliency measure, the more salient the contour. Although the weights for l, s, and c are arbitrary, it has been found that good results are obtained when ωl = ωs = ωc = 1/3. For the requirements of many computer vision applications, the three measures are given equal weights because one measure is normally not considered to be significantly more important than the others.
- After a saliency value is computed for each contour in the image, the contours are also ranked in decreasing order of saliency by saliency
value ranking unit 60. The most salient contours are then selected to be provided tovision application 65, which could be, for example, an object matching system. - Turning now to Figures 2a-e and 3a-e. Figures 2a-e involve an image of a flat bed truck obtained by a long wave forward-looking infra-red (FLIR) sensor, while Figures 3a-e relate to an image of a small communications van obtained in the visible spectrum by a TV sensor. Discussion about Figures 3a-e is identical to that which follows for Figures 2a-e.
- Figure 2a shows the gray level images received from
passive sensor 20. Figure 2b shows the Canny edges (edgels) obtained at small thresholds ofedge detector 20 to give a large number of edges. Figure 2c shows the most salient contours obtained fromsaliency value extractor 30. - Since corners are used for object recognition according to a preferred embodiment of the present invention, the performance of the salient contours extraction system and method of the present invention can be evaluated in terms of corners on and off the objects to be recognized. Figure 2d shows the cumulative number of on-target and off-target corners for the 50 most-salient contours. It can be seen that both the on-target and off-target curves are monotonically nondecreasing because they are cumulative curves. For better operation of
vision application 65, it is preferable to have more on-target corners and less off-target corners in the top few contours. Therefore, especially in the most salient contours, it is desirable to have the on-target curve above the off-target curve, with a wide gap between the two curves. It is also desirable to have the intersection between the two curves be high on the vertical axis and far to the right. The curves shown in Figure 2e depicts the ratio of the number of on-target corners to all image corners for the most salient contours. Studying the behavior of this ratio for the top few salient contours indicates the number of contours needed for vision applications such as object recognition. It is desirable for this curve to be high for the most salient contours and then drop for less salient ones. - The bar chart (Figure 4), corresponding to the gray level images of Figures 2a and 3a, shows the increase in the fraction of target corners obtained when only the most salient contours extracted according to the present invention are used. The darker bars show the fraction of target corners ascertained without
saliency value extractor 30, while the lighter bars reflect the fraction of target corners established by employingsaliency value extractor 30. - As mentioned earlier, the present invention is useful for a number of computer vision tasks, including object recognition. By separating an object from the background within an image, salient contours extraction becomes an important precursor to object recognition. Furthermore, since the clutter has been removed to a large extent, an object recognition system can proceed faster and more efficiently (i.e., more robust recognition).
- Returning to Figure 1, according to a preferred embodiment of the present invention, an object recognition application of
vision system 65 matches two-dimensional corners in an image, received from saliencyvalue combination unit 60, with the vertices of a three-dimensional model using an alignment technique.Object recognition system 65 first extracts corners in the image provided by saliencyvalue ranking unit 60. Then every triple of image corners is matched to every triple of three-dimensional model vertices to generate an alignment. The alignment is used to project the three-dimensional model to the two-dimensional image for verification. The salient contour extraction system and method of the present invention increases the ratio of image corners on the object to all image corners, thereby reducing the number of extraneous corners caused by image clutter, and resulting in a reduction of the search space (the combinations) in the corner matching stage of object recognition. Salient contours are also employed in a verification stage where only the salient image contours are used to verify the projected model contours. - Continuing now to Figures 5 and 6, it is apparent that the present invention can also be used in matching situations, such as motion and stereo. In motion matching situations (Figure 5), important for tracking objects within images, a plurality of image frames F₁,F₂,F₃,...Fn from a motion sequence are provided to
vision system 10 embodying the present invention.Vision system 10 extracts the most salient contours from frames F₁,F₂,F₃,...Fn and provides these contours to two dimensionalmatcher vision application 65. It should be noted that S₁,S₂,S₃,...Sn may comprise onevision system 10 or a plurality ofvision systems 10 operating in parallel.Matcher 65 then matches these contours to estimate motion parameters. The use of extracted salient contours makes it considerably easier formatcher 65 to match edges in frames of a motion sequence.Matcher 65 may then yield two-dimensional or three-dimensional motion estimation based on inferencing after the matching operation, may determine structure of the object of interest from its motion, may operate as a moving target indicator, etc. - Looking now at Figure 6, which illustrates an application of the present invention in a stereo matching situation, stereo image frames pair F₁,F₂ are provided to
vision system 10 embodying the present invention.Vision system 10 extracts the most salient contours from frames F₁,F₂ and provides these contours tomatcher vision application 65. Again, S₁,S₂ may comprise onevision system 10 or a plurality ofvision systems 10 operating in parallel.Matcher 65 then matches these contours to estimate depth and other three-dimensional information about the scene, based on inferencing. As in the previous matching situation, extracting salient contours according to the present invention makes matching much easier since clutter in the image is significantly reduced. - As an aside, the results achieved by use of the present invention are fairly independent of the kind of two-
dimensional sensor 15 used (e.g., TV, infra-red, etc.). Furthermore, the present invention operates primarily in a local context especially because an image may be occluded and/or there could be long, thin objects in the image. - It will be apparent to the skilled artisan that the present invention has numerous computer vision applications. Therefore, the present invention should not be limited to the computer vision applications discussed above, as these are provided for illustrative purposes only and not by way of limitation.
- While a specific embodiment of the invention has, been shown and described, various modifications and alternate embodiments will occur to those skilled in the art. Accordingly, it is intended that the invention be limited only in terms of the appended claims.
Note that only the average change of curvature is used and not total curvature. For this reason, the present invention does not require circles and very curved lines as the prior art requires for good results. Also, the system and method of the present invention is non-iterative and therefore does not have the convergence problems associated with the prior art.
Claims (20)
- A method for obtaining salient contours from two-dimensional images acquired by a sensor, comprising the steps of:
processing said two-dimensional images with an edge detector to produce edgels from each of said images;
linking said edgels into lists known as contours;
computing a saliency value for each of said contours;
ranking said contours in decreasing order of saliency; and
selecting predetermined ones from said ranked contours. - The method of Claim 1, wherein said edge detector is a Canny edge detector.
- The method of Claim 1, after said step of selecting, further comprising the step of discarding any ranked contours not selected.
- The method of Claim 1, wherein said step of computing a saliency value for each of said contours further comprises the steps of:
ascertaining each of a plurality of saliency measures;
applying a predetermined weight to each saliency measure; and
determining a saliency value based on said weighted saliency measures. - The method of Claim 4, wherein said step of ascertaining a plurality of saliency measures is performed in parallel.
- The method of Claim 4, wherein at least one of said plurality of saliency measures includes a smoothness saliency measure, S s, calculated as:
- The method of Claim 8, wherein said step of computing curvature employs a B-spline fitting.
- The method of Claim 8, wherein said step of determining any change of curvature is accomplished by convolving said computed curvature for each edgel using a difference operator.
- The method of Claim 4, wherein at least one of said plurality of saliency measures includes a length saliency measure, S l, calculated as:
- A system for object recognition, comprising:
a sensor for generating an image of a scene;
an edge detector connected to said sensor to detect edgels within said image;
a salient contour extractor connected to receive and process said detected edgels into contours; and
an object matching system connected to receive preselected ones of said processed contours and match them to known contours of predetermined objects. - The system of Claim 13, wherein said salient contour extractor comprises:
a linker which links said edgels into lists known as contours; and
a processing unit for determining a saliency value for each of said contours and ranking said contours in decreasing order of saliency. - The system of Claim 13, wherein said salient contour extractor comprises a plurality of contour extractors operating in parallel.
- The system of Claim 13, wherein said sensor is a FLIR.
- The system of Claim 13, wherein said sensor is a television camera.
- The system of Claim 13, wherein said edge detector is a Canny edge detector.
- The system of Claim 14, wherein said processing unit further comprises a computing unit for establishing a saliency value for each of said contours which determines a plurality of saliency measures for each saliency value.
- The system of Claim 19, wherein said plurality of saliency measures are preselectively weighted.
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
US676650 | 1991-03-28 | ||
US07/676,650 US5210799A (en) | 1991-03-28 | 1991-03-28 | System and method for ranking and extracting salient contours for target recognition |
Publications (3)
Publication Number | Publication Date |
---|---|
EP0506327A2 true EP0506327A2 (en) | 1992-09-30 |
EP0506327A3 EP0506327A3 (en) | 1994-11-23 |
EP0506327B1 EP0506327B1 (en) | 1998-08-12 |
Family
ID=24715385
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
EP92302501A Expired - Lifetime EP0506327B1 (en) | 1991-03-28 | 1992-03-24 | A system and method for ranking and extracting salient contours for target recognition |
Country Status (4)
Country | Link |
---|---|
US (2) | US5210799A (en) |
EP (1) | EP0506327B1 (en) |
JP (1) | JP3151284B2 (en) |
DE (1) | DE69226551T2 (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
EP0974931A1 (en) * | 1998-07-24 | 2000-01-26 | Xerox Corporation | Method and apparatus for identifying a plurality of sub-images in an input image |
WO2003030101A2 (en) * | 2001-10-03 | 2003-04-10 | Retinalyze Danmark A/S | Detection of vessels in an image |
CN112946598A (en) * | 2021-01-25 | 2021-06-11 | 西北工业大学 | Sky-wave radar ionosphere correction coefficient extraction method |
Families Citing this family (56)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP3034975B2 (en) * | 1991-03-26 | 2000-04-17 | 株式会社東芝 | Pattern feature extraction method |
US5325449A (en) * | 1992-05-15 | 1994-06-28 | David Sarnoff Research Center, Inc. | Method for fusing images and apparatus therefor |
US6278798B1 (en) * | 1993-08-09 | 2001-08-21 | Texas Instruments Incorporated | Image object recognition system and method |
US5434927A (en) * | 1993-12-08 | 1995-07-18 | Minnesota Mining And Manufacturing Company | Method and apparatus for machine vision classification and tracking |
EP0686932A3 (en) * | 1994-03-17 | 1997-06-25 | Texas Instruments Inc | A computer vision system to detect 3-D rectangular objects |
DE4413916A1 (en) * | 1994-04-21 | 1995-11-02 | Bodenseewerk Geraetetech | Passive friend / foe discrimination facility |
US6256529B1 (en) * | 1995-07-26 | 2001-07-03 | Burdette Medical Systems, Inc. | Virtual reality 3D visualization for surgical procedures |
AR000543A1 (en) * | 1995-12-26 | 1997-07-10 | Prignano Juan Pedro Alfr Volpe | Procedure to make three-dimensional figures from any image expressed on a flat surface |
AU1689697A (en) * | 1995-12-26 | 1997-07-17 | Juan Pedro Alfredo Hector Volpe Prignano | A method for creating three-dimensional figures or forms from any flat surface image |
US5850469A (en) * | 1996-07-09 | 1998-12-15 | General Electric Company | Real time tracking of camera pose |
US5859891A (en) | 1997-03-07 | 1999-01-12 | Hibbard; Lyn | Autosegmentation/autocontouring system and method for use with three-dimensional radiation therapy treatment planning |
US6084590A (en) * | 1997-04-07 | 2000-07-04 | Synapix, Inc. | Media production with correlation of image stream and abstract objects in a three-dimensional virtual stage |
US6160907A (en) * | 1997-04-07 | 2000-12-12 | Synapix, Inc. | Iterative three-dimensional process for creating finished media content |
US6124864A (en) * | 1997-04-07 | 2000-09-26 | Synapix, Inc. | Adaptive modeling and segmentation of visual image streams |
US5983218A (en) * | 1997-06-30 | 1999-11-09 | Xerox Corporation | Multimedia database for use over networks |
US6185316B1 (en) | 1997-11-12 | 2001-02-06 | Unisys Corporation | Self-authentication apparatus and method |
CA2333583C (en) | 1997-11-24 | 2005-11-08 | Everette C. Burdette | Real time brachytherapy spatial registration and visualization system |
US6094508A (en) * | 1997-12-08 | 2000-07-25 | Intel Corporation | Perceptual thresholding for gradient-based local edge detection |
US6266053B1 (en) | 1998-04-03 | 2001-07-24 | Synapix, Inc. | Time inheritance scene graph for representation of media content |
US6297825B1 (en) | 1998-04-06 | 2001-10-02 | Synapix, Inc. | Temporal smoothing of scene analysis data for image sequence generation |
US6249285B1 (en) | 1998-04-06 | 2001-06-19 | Synapix, Inc. | Computer assisted mark-up and parameterization for scene analysis |
US6535639B1 (en) * | 1999-03-12 | 2003-03-18 | Fuji Xerox Co., Ltd. | Automatic video summarization using a measure of shot importance and a frame-packing method |
US6983083B2 (en) * | 2001-02-13 | 2006-01-03 | Eastman Kodak Company | Image specific perceived overall contrast prediction |
KR100374708B1 (en) * | 2001-03-06 | 2003-03-04 | 에버미디어 주식회사 | Non-contact type human iris recognition method by correction of rotated iris image |
US20020154833A1 (en) * | 2001-03-08 | 2002-10-24 | Christof Koch | Computation of intrinsic perceptual saliency in visual environments, and applications |
US7438685B2 (en) | 2001-11-05 | 2008-10-21 | Computerized Medical Systems, Inc. | Apparatus and method for registration, guidance and targeting of external beam radiation therapy |
JP4187448B2 (en) * | 2002-03-07 | 2008-11-26 | 富士通マイクロエレクトロニクス株式会社 | Method and apparatus for tracking moving object in image |
US7187800B2 (en) | 2002-08-02 | 2007-03-06 | Computerized Medical Systems, Inc. | Method and apparatus for image segmentation using Jensen-Shannon divergence and Jensen-Renyi divergence |
US20040161153A1 (en) * | 2003-02-18 | 2004-08-19 | Michael Lindenbaum | Context-based detection of structured defects in an image |
US7376252B2 (en) * | 2003-11-25 | 2008-05-20 | Ge Medical Systems Global Technology Company, Llc | User interactive method and user interface for detecting a contour of an object |
GB2415562B (en) * | 2004-06-23 | 2007-11-21 | Hewlett Packard Development Co | Image processing |
US7738705B2 (en) * | 2004-06-30 | 2010-06-15 | Stefano Casadei | Hierarchical method and system for pattern recognition and edge detection |
US7463753B2 (en) * | 2004-09-15 | 2008-12-09 | Raytheon Company | FLIR-to-missile boresight correlation and non-uniformity compensation of the missile seeker |
US7650030B2 (en) * | 2004-12-03 | 2010-01-19 | Sarnoff Corporation | Method and apparatus for unsupervised learning of discriminative edge measures for vehicle matching between non-overlapping cameras |
GB2422739B (en) * | 2005-01-31 | 2010-07-14 | Hewlett Packard Development Co | Image processing method and apparatus |
US7860301B2 (en) * | 2005-02-11 | 2010-12-28 | Macdonald Dettwiler And Associates Inc. | 3D imaging system |
US7613363B2 (en) * | 2005-06-23 | 2009-11-03 | Microsoft Corp. | Image superresolution through edge extraction and contrast enhancement |
WO2007108056A1 (en) * | 2006-03-16 | 2007-09-27 | Fujitsu Limited | Video image recognition system and video image recognition program |
KR100762670B1 (en) * | 2006-06-07 | 2007-10-01 | 삼성전자주식회사 | Method and device for generating disparity map from stereo image and stereo matching method and device therefor |
US7940955B2 (en) * | 2006-07-26 | 2011-05-10 | Delphi Technologies, Inc. | Vision-based method of determining cargo status by boundary detection |
US8150101B2 (en) | 2006-11-13 | 2012-04-03 | Cybernet Systems Corporation | Orientation invariant object identification using model-based image processing |
JP2008269471A (en) * | 2007-04-24 | 2008-11-06 | Sony Corp | Similar image decision device, similar image decision method, program, and recording medium |
TWI394095B (en) * | 2008-10-22 | 2013-04-21 | Ind Tech Res Inst | Image detecting method and system thereof |
US8542950B2 (en) * | 2009-06-02 | 2013-09-24 | Yahoo! Inc. | Finding iconic images |
US8781160B2 (en) * | 2009-12-31 | 2014-07-15 | Indian Institute Of Technology Bombay | Image object tracking and segmentation using active contours |
WO2011152893A1 (en) * | 2010-02-10 | 2011-12-08 | California Institute Of Technology | Methods and systems for generating saliency models through linear and/or nonlinear integration |
US9070182B1 (en) | 2010-07-13 | 2015-06-30 | Google Inc. | Method and system for automatically cropping images |
US8363984B1 (en) | 2010-07-13 | 2013-01-29 | Google Inc. | Method and system for automatically cropping images |
CN103700091B (en) * | 2013-12-01 | 2016-08-31 | 北京航空航天大学 | Based on the image significance object detection method that multiple dimensioned low-rank decomposition and structural information are sensitive |
US9269018B2 (en) | 2014-01-14 | 2016-02-23 | Microsoft Technology Licensing, Llc | Stereo image processing using contours |
US9299007B2 (en) * | 2014-01-28 | 2016-03-29 | Ncr Corporation | Methods and apparatus for item identification using brightness compensation |
US9563953B2 (en) * | 2014-08-28 | 2017-02-07 | Qualcomm Incorporated | Systems and methods for determining a seam |
US9626584B2 (en) * | 2014-10-09 | 2017-04-18 | Adobe Systems Incorporated | Image cropping suggestion using multiple saliency maps |
CN104992144B (en) * | 2015-06-11 | 2018-05-29 | 电子科技大学 | The differentiating method of power transmission line and highway in remote sensing images |
CN104966065B (en) * | 2015-06-23 | 2018-11-09 | 电子科技大学 | target identification method and device |
US10713792B1 (en) * | 2017-01-13 | 2020-07-14 | Amazon Technologies, Inc. | System and apparatus for image processing |
Family Cites Families (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
DE3274760D1 (en) * | 1981-10-01 | 1987-01-29 | Commw Of Australia | PHOTOGRAMMETRIC STEREOPLOTTER |
US4648120A (en) * | 1982-07-02 | 1987-03-03 | Conoco Inc. | Edge and line detection in multidimensional noisy, imagery data |
JPS59182688A (en) * | 1983-03-31 | 1984-10-17 | Toshiba Corp | Stereoscopic processor |
IL70213A (en) * | 1983-11-13 | 1988-02-29 | Paul Fenster | Digital fluorographic image enhancement system |
DE3687760T2 (en) * | 1985-03-13 | 1993-06-09 | Topcon Corp | DEVICE AND METHOD FOR MEASURING COORDINATES. |
US4817166A (en) * | 1986-05-05 | 1989-03-28 | Perceptics Corporation | Apparatus for reading a license plate |
US4901362A (en) * | 1988-08-08 | 1990-02-13 | Raytheon Company | Method of recognizing patterns |
-
1991
- 1991-03-28 US US07/676,650 patent/US5210799A/en not_active Expired - Lifetime
-
1992
- 1992-03-24 DE DE69226551T patent/DE69226551T2/en not_active Expired - Fee Related
- 1992-03-24 EP EP92302501A patent/EP0506327B1/en not_active Expired - Lifetime
- 1992-03-27 JP JP07165992A patent/JP3151284B2/en not_active Expired - Fee Related
-
1995
- 1995-06-07 US US08/483,351 patent/US5566246A/en not_active Expired - Lifetime
Non-Patent Citations (1)
Title |
---|
COMPUTER VISION GRAPHICS AND IMAGE PROCESSING, vol.46, no.2, May 1989, DULUTH, MA US pages 162 - 174 BEGHDADI ET AL. 'Contrast Enhancement Technique Based on Local Detection of Edges' * |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
EP0974931A1 (en) * | 1998-07-24 | 2000-01-26 | Xerox Corporation | Method and apparatus for identifying a plurality of sub-images in an input image |
WO2003030101A2 (en) * | 2001-10-03 | 2003-04-10 | Retinalyze Danmark A/S | Detection of vessels in an image |
WO2003030101A3 (en) * | 2001-10-03 | 2004-03-25 | Retinalyze As | Detection of vessels in an image |
CN112946598A (en) * | 2021-01-25 | 2021-06-11 | 西北工业大学 | Sky-wave radar ionosphere correction coefficient extraction method |
CN112946598B (en) * | 2021-01-25 | 2022-08-16 | 西北工业大学 | Sky-wave radar ionosphere correction coefficient extraction method |
Also Published As
Publication number | Publication date |
---|---|
JP3151284B2 (en) | 2001-04-03 |
DE69226551D1 (en) | 1998-09-17 |
EP0506327A3 (en) | 1994-11-23 |
DE69226551T2 (en) | 1998-12-24 |
JPH06124344A (en) | 1994-05-06 |
US5566246A (en) | 1996-10-15 |
US5210799A (en) | 1993-05-11 |
EP0506327B1 (en) | 1998-08-12 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US5210799A (en) | System and method for ranking and extracting salient contours for target recognition | |
EP1318477B1 (en) | Robust appearance models for visual motion analysis and tracking | |
JP2773818B2 (en) | Automatic image recognition apparatus and method | |
US8385630B2 (en) | System and method of processing stereo images | |
US6516087B1 (en) | Method for real time correlation of stereo images | |
Elder et al. | The statistics of natural image contours | |
CN109754009B (en) | Article identification method, article identification device, vending system and storage medium | |
KR20030045098A (en) | Balanced object tracker in an image sequence | |
CN112070782B (en) | Method, device, computer readable medium and electronic equipment for identifying scene contour | |
EP3352138A1 (en) | Method and apparatus for processing a 3d scene | |
CN113971751A (en) | Training feature extraction model, and method and device for detecting similar images | |
CN108305267B (en) | Object segmentation method, device, apparatus, storage medium, and program | |
CN112561879A (en) | Ambiguity evaluation model training method, image ambiguity evaluation method and device | |
CN109255792A (en) | A kind of dividing method of video image, device, terminal device and storage medium | |
CN113592706B (en) | Method and device for adjusting homography matrix parameters | |
CN115690545A (en) | Training target tracking model and target tracking method and device | |
Stentiford | Attention-based vanishing point detection | |
CN115631370A (en) | Identification method and device of MRI (magnetic resonance imaging) sequence category based on convolutional neural network | |
CN114445689A (en) | Multi-scale weighted fusion target detection method and system guided by target prior information | |
JPH11120351A (en) | Image matching device and storage medium to store image matching program | |
Rao | Extracting salient contours for target recognition: algorithm and performance evaluation | |
JP3047952B2 (en) | Image processing device | |
CN113570667B (en) | Visual inertial navigation compensation method and device and storage medium | |
Menard et al. | Adaptive stereo matching in correlation scale-space | |
CN116708995B (en) | Photographic composition method, photographic composition device and photographic equipment |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PUAI | Public reference made under article 153(3) epc to a published international application that has entered the european phase |
Free format text: ORIGINAL CODE: 0009012 |
|
AK | Designated contracting states |
Kind code of ref document: A2 Designated state(s): DE FR GB IT NL |
|
PUAL | Search report despatched |
Free format text: ORIGINAL CODE: 0009013 |
|
AK | Designated contracting states |
Kind code of ref document: A3 Designated state(s): DE FR GB IT NL |
|
17P | Request for examination filed |
Effective date: 19950428 |
|
17Q | First examination report despatched |
Effective date: 19970220 |
|
GRAG | Despatch of communication of intention to grant |
Free format text: ORIGINAL CODE: EPIDOS AGRA |
|
GRAG | Despatch of communication of intention to grant |
Free format text: ORIGINAL CODE: EPIDOS AGRA |
|
GRAH | Despatch of communication of intention to grant a patent |
Free format text: ORIGINAL CODE: EPIDOS IGRA |
|
GRAH | Despatch of communication of intention to grant a patent |
Free format text: ORIGINAL CODE: EPIDOS IGRA |
|
GRAA | (expected) grant |
Free format text: ORIGINAL CODE: 0009210 |
|
ITF | It: translation for a ep patent filed |
Owner name: BARZANO' E ZANARDO ROMA S.P.A. |
|
AK | Designated contracting states |
Kind code of ref document: B1 Designated state(s): DE FR GB IT NL |
|
REF | Corresponds to: |
Ref document number: 69226551 Country of ref document: DE Date of ref document: 19980917 |
|
ET | Fr: translation filed | ||
PLBE | No opposition filed within time limit |
Free format text: ORIGINAL CODE: 0009261 |
|
STAA | Information on the status of an ep patent application or granted ep patent |
Free format text: STATUS: NO OPPOSITION FILED WITHIN TIME LIMIT |
|
26N | No opposition filed | ||
REG | Reference to a national code |
Ref country code: GB Ref legal event code: IF02 |
|
PGFP | Annual fee paid to national office [announced via postgrant information from national office to epo] |
Ref country code: GB Payment date: 20070202 Year of fee payment: 16 |
|
PGFP | Annual fee paid to national office [announced via postgrant information from national office to epo] |
Ref country code: DE Payment date: 20070330 Year of fee payment: 16 |
|
PGFP | Annual fee paid to national office [announced via postgrant information from national office to epo] |
Ref country code: IT Payment date: 20070628 Year of fee payment: 16 |
|
PGFP | Annual fee paid to national office [announced via postgrant information from national office to epo] |
Ref country code: FR Payment date: 20070301 Year of fee payment: 16 |
|
PGFP | Annual fee paid to national office [announced via postgrant information from national office to epo] |
Ref country code: NL Payment date: 20080219 Year of fee payment: 17 |
|
GBPC | Gb: european patent ceased through non-payment of renewal fee |
Effective date: 20080324 |
|
REG | Reference to a national code |
Ref country code: FR Ref legal event code: ST Effective date: 20081125 |
|
PG25 | Lapsed in a contracting state [announced via postgrant information from national office to epo] |
Ref country code: DE Free format text: LAPSE BECAUSE OF NON-PAYMENT OF DUE FEES Effective date: 20081001 |
|
PG25 | Lapsed in a contracting state [announced via postgrant information from national office to epo] |
Ref country code: FR Free format text: LAPSE BECAUSE OF NON-PAYMENT OF DUE FEES Effective date: 20080331 |
|
PG25 | Lapsed in a contracting state [announced via postgrant information from national office to epo] |
Ref country code: GB Free format text: LAPSE BECAUSE OF NON-PAYMENT OF DUE FEES Effective date: 20080324 |
|
PG25 | Lapsed in a contracting state [announced via postgrant information from national office to epo] |
Ref country code: IT Free format text: LAPSE BECAUSE OF NON-PAYMENT OF DUE FEES Effective date: 20080324 |
|
NLV4 | Nl: lapsed or anulled due to non-payment of the annual fee |
Effective date: 20091001 |
|
PG25 | Lapsed in a contracting state [announced via postgrant information from national office to epo] |
Ref country code: NL Free format text: LAPSE BECAUSE OF NON-PAYMENT OF DUE FEES Effective date: 20091001 |